Skip to content

isc: two-sample permutation test fails for pairwise ISCs #449

Closed

Description

@snastase

When computing pairwise ISCs, the output is a condensed triangle vector (via squareform) with length N * (N - 1) / 2, where N is the number of subjects. In the context of two separate (non-paired) groups, ISCs are computed separately and the resulting condensed vectors are v-stacked (note that there may not be the same number of subjects in each group). When passing this stacked pairwise ISC array for two groups to permutation_isc, the _check_isc_input doesn't handle the stacked groups properly. It assumes there's one group and breaks because that stacked array likely won't have the correct number of elements for squareform. I think most of the rest of the permutation test handles this situation properly, but the logic of _check_isc_input needs to be adjusted, and we need a test confirms this is handled properly. Thanks to @mcdoar9 for catching this!

Activity

  1. self-assigned this
    on Dec 8, 2019
  2. snastase commented on Jul 31, 2020

    @snastase
    ContributorAuthor

    We've confirmed that this is not really a bug; more that the documentation is unclear as to how two-sample pairwise ISCs are to be supplied to permutation_isc. PR #476 should clarify this, so I'm closing this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions